Jamie R. Sykes, Katherine J. Denby, Daniel W. Franks
{"title":"Improving computer vision for plant pathology through advanced training techniques","authors":"Jamie R. Sykes, Katherine J. Denby, Daniel W. Franks","doi":"10.1002/aps3.70010","DOIUrl":null,"url":null,"abstract":"<div>\n \n \n <section>\n \n <h3> Premise</h3>\n \n <p>This study investigates advanced training techniques to improve the performance of convolutional neural networks for disease detection in cocoa, <i>Theobroma cacao</i>.</p>\n </section>\n \n <section>\n \n <h3> Methods</h3>\n \n <p>Despite recent stagnation in accuracy improvements in computer vision for image classification, our research demonstrates significant advancements in performance through semi-supervised learning, specialised loss functions, and the inclusion of a non-cocoa class.</p>\n </section>\n \n <section>\n \n <h3> Results</h3>\n \n <p>Semi-supervised learning reduced overfitting and enhanced generalisability, particularly for subtle symptoms. The non-cocoa class exposed models to a broad range of relevant features, significantly improving model robustness and performance in difficult cases. Grad-CAM for qualitative assessment provided valuable insights into model behaviour, highlighting cases of overfitting missed by summary statistics. We also describe dynamic focal loss, a novel loss function that uses an empirical measure of difficulty to weight each image. Our results suggest that while PhytNet shows promise in terms of computational efficiency and superior handling of difficult images, ResNet18 with semi-supervised learning and dynamic focal loss emerged as the strongest contender for real-world deployment.</p>\n </section>\n \n <section>\n \n <h3> Discussion</h3>\n \n <p>This research underscores the potential of semi-supervised learning and advanced loss functions in enhancing the applicability of deep learning models in agricultural disease management. It also presents a new high-quality benchmark dataset of 7220 images of diseased and healthy cocoa trees, offering a much greater and more realistic challenge than the Plan Village dataset.</p>\n </section>\n </div>","PeriodicalId":8022,"journal":{"name":"Applications in Plant Sciences","volume":"13 3","pages":""},"PeriodicalIF":2.4000,"publicationDate":"2025-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/aps3.70010","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Applications in Plant Sciences","FirstCategoryId":"99","ListUrlMain":"https://onlinelibrary.wiley.com/doi/10.1002/aps3.70010","RegionNum":3,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"PLANT SCIENCES","Score":null,"Total":0}
引用次数: 0
Abstract
Premise
This study investigates advanced training techniques to improve the performance of convolutional neural networks for disease detection in cocoa, Theobroma cacao.
Methods
Despite recent stagnation in accuracy improvements in computer vision for image classification, our research demonstrates significant advancements in performance through semi-supervised learning, specialised loss functions, and the inclusion of a non-cocoa class.
Results
Semi-supervised learning reduced overfitting and enhanced generalisability, particularly for subtle symptoms. The non-cocoa class exposed models to a broad range of relevant features, significantly improving model robustness and performance in difficult cases. Grad-CAM for qualitative assessment provided valuable insights into model behaviour, highlighting cases of overfitting missed by summary statistics. We also describe dynamic focal loss, a novel loss function that uses an empirical measure of difficulty to weight each image. Our results suggest that while PhytNet shows promise in terms of computational efficiency and superior handling of difficult images, ResNet18 with semi-supervised learning and dynamic focal loss emerged as the strongest contender for real-world deployment.
Discussion
This research underscores the potential of semi-supervised learning and advanced loss functions in enhancing the applicability of deep learning models in agricultural disease management. It also presents a new high-quality benchmark dataset of 7220 images of diseased and healthy cocoa trees, offering a much greater and more realistic challenge than the Plan Village dataset.
期刊介绍:
Applications in Plant Sciences (APPS) is a monthly, peer-reviewed, open access journal promoting the rapid dissemination of newly developed, innovative tools and protocols in all areas of the plant sciences, including genetics, structure, function, development, evolution, systematics, and ecology. Given the rapid progress today in technology and its application in the plant sciences, the goal of APPS is to foster communication within the plant science community to advance scientific research. APPS is a publication of the Botanical Society of America, originating in 2009 as the American Journal of Botany''s online-only section, AJB Primer Notes & Protocols in the Plant Sciences.
APPS publishes the following types of articles: (1) Protocol Notes describe new methods and technological advancements; (2) Genomic Resources Articles characterize the development and demonstrate the usefulness of newly developed genomic resources, including transcriptomes; (3) Software Notes detail new software applications; (4) Application Articles illustrate the application of a new protocol, method, or software application within the context of a larger study; (5) Review Articles evaluate available techniques, methods, or protocols; (6) Primer Notes report novel genetic markers with evidence of wide applicability.